International Journal of Epidemiology 2000;29:807-812
© International Epidemiological Association 2000
Selective screening for cervical neoplasia: an approach for resource-poor settings
a Departments of Epidemiology, and
b Medicine, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA.
c Family Health International, Research Triangle Park, NC, USA.
Reprint requests to: William C Miller, Department of Epidemiology, 2105F McGavran-Greenberg Building, CB # 7400, UNC—Chapel Hill, Chapel Hill, NC 27599–7400, USA.
Background Cervical malignancies are the leading cause of cancer-related morbidity and mortality among women in developing countries. Although early detection programmes using cytological methods, followed by aggressive treatment of precursor lesions are accepted as the main disease control strategy, fiscal limitations make this strategy unfeasible in many countries.
Methods To screen selectively, we developed two risk scores using data from a population-based case-control study in Jamaica with 202 cases and 363 controls. Independent risk factors for cervical neoplasia were determined using logistic regression. An unweighted risk score for each subject was developed by a simple count of risk factors present and a weighted risk score was calculated by summing regression coefficients for each risk factor.
Results Four patient characteristics were independently predictive of cervical neoplasia, older age (OR = 3.4, 95% CI : 1.8–6.7), 
4 pregnancies (OR = 5.6, 95% CI : 1.2– 18.7), poverty (OR = 2.1, 95% CI : 1.3–3.3) and cigarette smoking (OR = 1.9, 95% CI : 1.2–3.2). Using cut-off points of 20 for the weighted scores and >3 for unweighted scores, the sensitivity and specificity were 65% and 69% for the unweighted score and 75% and 61%, respectively, for the weighted score. Areas under the receiver operating characteristic (ROC) curves for the weighted versus the unweighted scores were similar, suggesting similar overall accuracy.
Conclusion Selective screening using risk assessment strategies is potentially useful, particularly in resource-poor settings. However, whether weighting factors is essential is dependent on prevalence of factors in a given setting. Although this approach needs validation in other populations, women at highest risk for cervical neoplasia can be identified using demographic factors available during a regular clinic visit.
Keywords Cervical neoplasia, screening, risk scores, receiver operating characteristic (ROC) curves
Accepted 7 February 2000